Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Claude Fable 5.2 website refactor interrupted by usage limit

AI summary

In a first-person test, the author used Claude Code with Claude Fable 5.2 to orchestrate multiple agents, develop three homepage directions and modify 20 files for the ShipSite website. The session ran for 1 hour and 25 minutes and cost $52.55, but the account's five-hour usage limit interrupted final verification. The author later ran the remaining checks manually and found that four mobile pages became 0.1 to 0.3 seconds slower. The report also states that Anthropic had announced faster generation, lower token prices and new sandbox-escape evaluations.

In-depth insight

The test documents the cost, quota exposure and incomplete final checks involved in a long-running, multi-agent website refactor. It also supports the author's recommendation to split direction selection and implementation into separate stages.

做独立开发出海,最耗精力的往往是给产品做官网门面:调配色、对设计、挨个修手机端自适应,一折腾就是好几天。

昨天,Anthropic 刚发布了新模型 Claude Fable 5.2(2026年11月9日发布,见 Anthropic 官网公告及开发文档)。官方对它的定位是专打高要求推理与数小时长任务编排;相比上一代,它的生成速度快了最多 30%,输入 8/输出8 / 输出 8/输出40 每百万 token 的价格也更低,并引入了防沙盒逃逸(sandbox escapes)的新安全评估。

正好我自己的网站生成器项目 ShipSite 官网一直想翻新,我就拿它试了一把:在终端里跑 Claude Code 命令行,把思考档位拉满,只发了一句话,告诉它这个代码库归它,让它向 Awwwards 获奖站看齐自主重构,中途别问我。

它在后台跑了 85 分钟,派生多个子 Agent 做了 3 个首页方向,选定「图志」方案后改了 20 个文件、写了 4320 行代码(/cost 的统计,含它自己的工具脚本);但在最后阶段撞上了 5 小时配额上限,实测提示「你停止了操作」,账单花了 52.55 美元。

#Anthropic#Claude Fable 5.2#Claude Code#ShipSite#Haiku 5.5

Original · 10 min readRead the original
Group chat

No comments yet. Start the conversation.